A Study on Noise Reduction Method Based on Noise Reconstruction System with Step Size Control

A noise reduction method is proposed to reduce background noise in noisy speech. We have investigated a noise reconstruction system using a linear prediction error filter (LPEF) and a noise reconstruction filter (NRF). An input signal of a LPEF becomes a white signal. Assuming that background noise is generated by exciting a linear system by a white signal, the background noise can be reconstructed from white noise by estimating the linear system. The NRF estimates the linear system. However, since residual speech is included in an input signal of a NRF, it is necessary to use a small step size, which enables us to update coefficients of a NRF without estimating a speech signal. Therefore, it becomes difficult for a NRF to track non-stationary noise. In order to solve the problem, a noise reconstruction system using step size control is proposed.